Abstract: Based on adaptive morphological filtering,a defect extraction method is proposed in this paper to meet needs of machine vision automatic identification system of the arc magnet's surface.To extract defects with low contrast,textured background and uneven brightness,the method presents a novel adaptive morphological filter.Sub-regional and progressive scan was executed to estimate maximum size of defects on each line according to gray level variety in different regions of the image.The one-dimensional rod-like structural elements of each morphological filter could be adjusted automatically with maximum size of defects.The defects are removed or weakened by progressive and adaptive morphological filtering,and the image background(threshold curve) was obtained.Then defects in the inspected surface could be extracted by comparing the original image with the background image.The experimental results indicate that the proposed method can extract defects of different sub-regions rapidly and accurately with good versatility,and can be used for online automatic identification to defects.